We are not able to identify AI-generated images
Adrien Pavão
TL;DR
This paper investigates whether humans can reliably differentiate AI-generated portraits from real photographs in an online setting. It uses an interactive 20-image task with 120 challenging cases (60 real from CC12M, 60 AI-generated via MidJourney v7) and collects data from 165 participants across 233 sessions. The main finding is that average accuracy is about 54%, only slightly above chance, with longer viewing times yielding better discrimination and substantial variation across images; improvements from repeated sessions are limited. The results highlight the inadequacy of relying on human judgment alone to detect synthetic imagery and call for increased awareness, ethical guidelines, and the development of robust detection tools, while suggesting future work to broaden modalities and include confidence measures.
Abstract
AI-generated images are now pervasive online, yet many people believe they can easily tell them apart from real photographs. We test this assumption through an interactive web experiment where participants classify 20 images as real or AI-generated. Our dataset contains 120 difficult cases: real images sampled from CC12M, and carefully curated AI-generated counterparts produced with MidJourney. In total, 165 users completed 233 sessions. Their average accuracy was 54%, only slightly above random guessing, with limited improvement across repeated attempts. Response times averaged 7.3 seconds, and some images were consistently more deceptive than others. These results indicate that, even on relatively simple portrait images, humans struggle to reliably detect AI-generated content. As synthetic media continues to improve, human judgment alone is becoming insufficient for distinguishing real from artificial data. These findings highlight the need for greater awareness and ethical guidelines as AI-generated media becomes increasingly indistinguishable from reality.